Re: [PATCH 2/2] - usbserial: race-condition fix.

[Eduardo Pereira Habkost]

On Wed, Nov 23, 2005 at 09:36:55AM -0200, Luiz Fernando Capitulino wrote:
> On Tue, 22 Nov 2005 14:13:53 -0800
> Greg KH <gregkh@xxxxxxx> wrote:
> 
> | On Tue, Nov 22, 2005 at 07:59:26PM -0200, Luiz Fernando Capitulino wrote:
> | > @@ -60,6 +61,7 @@ struct usb_serial_port {
> | >  	struct usb_serial *	serial;
> | >  	struct tty_struct *	tty;
> | >  	spinlock_t		lock;
> | > +	struct semaphore        sem;
> | 
> | You forgot to document what this semaphore is used for.
> 
>  Okay.
> 
> | Hm, can we just use the spinlock already present in the port structure
> | for this?  Well, drop the spinlock and use the semaphore?  Yeah, that
> | means grabbing a semaphore for ever write for some devices, but USB data
> | rates are slow enough it wouldn't matter :)
> 
>  As far as I read the code, I found that spinlock is only used by the
> generic driver, in the
> drivers/usb/serial/generic.c:usb_serial_generic_write() function.
> 
>  Can we drop the spinlock there and use our new semaphore? Or should we
> create a new spinlock just to use there?

The spin_lock is used only to protect write_urb_busy. An atomic_t seem
to be more appropriate for it. If we do that, I guess we can remove the
(then unused) spinlock.

So we have three proposed changes:

- Add semaphore to serialize close()/open() (properly documented)
- Replace write_urb_busy with an atomic_t
- Remove the spinlock

> 
>  I ask it because the semaphore will be used to serialize open()/close()
> operations in the usb-serial driver, is a bit weird to use the same
> semaphore in a write() function of other driver.

I agree.

-- 
Eduardo

Attachment: pgp00000.pgp
Description: PGP signature